Commit | Line | Data |
---|---|---|
f87e4cac JF |
1 | /* |
2 | * Xen SMP support | |
3 | * | |
4 | * This file implements the Xen versions of smp_ops. SMP under Xen is | |
5 | * very straightforward. Bringing a CPU up is simply a matter of | |
6 | * loading its initial context and setting it running. | |
7 | * | |
8 | * IPIs are handled through the Xen event mechanism. | |
9 | * | |
10 | * Because virtual CPUs can be scheduled onto any real CPU, there's no | |
11 | * useful topology information for the kernel to make use of. As a | |
12 | * result, all CPUs are treated as if they're single-core and | |
13 | * single-threaded. | |
f87e4cac JF |
14 | */ |
15 | #include <linux/sched.h> | |
16 | #include <linux/err.h> | |
5a0e3ad6 | 17 | #include <linux/slab.h> |
f87e4cac | 18 | #include <linux/smp.h> |
1ff2b0c3 | 19 | #include <linux/irq_work.h> |
466318a8 | 20 | #include <linux/tick.h> |
f87e4cac JF |
21 | |
22 | #include <asm/paravirt.h> | |
23 | #include <asm/desc.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/cpu.h> | |
26 | ||
27 | #include <xen/interface/xen.h> | |
28 | #include <xen/interface/vcpu.h> | |
29 | ||
30 | #include <asm/xen/interface.h> | |
31 | #include <asm/xen/hypercall.h> | |
32 | ||
ea5b8f73 | 33 | #include <xen/xen.h> |
f87e4cac JF |
34 | #include <xen/page.h> |
35 | #include <xen/events.h> | |
36 | ||
ed467e69 | 37 | #include <xen/hvc-console.h> |
f87e4cac JF |
38 | #include "xen-ops.h" |
39 | #include "mmu.h" | |
a2ef5dc2 | 40 | #include "smp.h" |
f87e4cac | 41 | |
b78936e1 | 42 | cpumask_var_t xen_cpu_initialized_map; |
f87e4cac | 43 | |
9547689f KRW |
44 | struct xen_common_irq { |
45 | int irq; | |
46 | char *name; | |
47 | }; | |
ee336e10 KRW |
48 | static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 }; |
49 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 }; | |
50 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 }; | |
51 | static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 }; | |
9547689f | 52 | static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 }; |
f87e4cac JF |
53 | |
54 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id); | |
3b16cf87 | 55 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id); |
1ff2b0c3 | 56 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id); |
f87e4cac JF |
57 | |
58 | /* | |
184748cc | 59 | * Reschedule call back. |
f87e4cac JF |
60 | */ |
61 | static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id) | |
62 | { | |
1b437c8c | 63 | inc_irq_stat(irq_resched_count); |
184748cc | 64 | scheduler_ipi(); |
38bb5ab4 | 65 | |
f87e4cac JF |
66 | return IRQ_HANDLED; |
67 | } | |
68 | ||
148f9bb8 | 69 | static void cpu_bringup(void) |
f87e4cac | 70 | { |
e8c9e788 | 71 | int cpu; |
f87e4cac JF |
72 | |
73 | cpu_init(); | |
d68d82af | 74 | touch_softlockup_watchdog(); |
c7b75947 JF |
75 | preempt_disable(); |
76 | ||
5840c84b MR |
77 | /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */ |
78 | if (!xen_feature(XENFEAT_supervisor_mode_kernel)) { | |
79 | xen_enable_sysenter(); | |
80 | xen_enable_syscall(); | |
81 | } | |
c7b75947 JF |
82 | cpu = smp_processor_id(); |
83 | smp_store_cpu_info(cpu); | |
84 | cpu_data(cpu).x86_max_cores = 1; | |
85 | set_cpu_sibling_map(cpu); | |
f87e4cac JF |
86 | |
87 | xen_setup_cpu_clockevents(); | |
88 | ||
106b4438 KRW |
89 | notify_cpu_starting(cpu); |
90 | ||
d7d3756c | 91 | set_cpu_online(cpu, true); |
106b4438 | 92 | |
2a442c9c | 93 | cpu_set_state_online(cpu); /* Implies full memory barrier. */ |
c7b75947 | 94 | |
f87e4cac JF |
95 | /* We can take interrupts now: we're officially "up". */ |
96 | local_irq_enable(); | |
d68d82af AN |
97 | } |
98 | ||
a2ef5dc2 MR |
99 | /* |
100 | * Note: cpu parameter is only relevant for PVH. The reason for passing it | |
101 | * is we can't do smp_processor_id until the percpu segments are loaded, for | |
102 | * which we need the cpu number! So we pass it in rdi as first parameter. | |
103 | */ | |
104 | asmlinkage __visible void cpu_bringup_and_idle(int cpu) | |
d68d82af | 105 | { |
a2ef5dc2 | 106 | #ifdef CONFIG_XEN_PVH |
5840c84b MR |
107 | if (xen_feature(XENFEAT_auto_translated_physmap) && |
108 | xen_feature(XENFEAT_supervisor_mode_kernel)) | |
c9f6e997 | 109 | xen_pvh_secondary_vcpu_init(cpu); |
5840c84b | 110 | #endif |
d68d82af | 111 | cpu_bringup(); |
7d1a9417 | 112 | cpu_startup_entry(CPUHP_ONLINE); |
f87e4cac JF |
113 | } |
114 | ||
53b94fdc KRW |
115 | static void xen_smp_intr_free(unsigned int cpu) |
116 | { | |
ee336e10 | 117 | if (per_cpu(xen_resched_irq, cpu).irq >= 0) { |
9547689f | 118 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL); |
ee336e10 | 119 | per_cpu(xen_resched_irq, cpu).irq = -1; |
b85fffec KRW |
120 | kfree(per_cpu(xen_resched_irq, cpu).name); |
121 | per_cpu(xen_resched_irq, cpu).name = NULL; | |
ee336e10 KRW |
122 | } |
123 | if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) { | |
9547689f | 124 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL); |
ee336e10 | 125 | per_cpu(xen_callfunc_irq, cpu).irq = -1; |
b85fffec KRW |
126 | kfree(per_cpu(xen_callfunc_irq, cpu).name); |
127 | per_cpu(xen_callfunc_irq, cpu).name = NULL; | |
ee336e10 KRW |
128 | } |
129 | if (per_cpu(xen_debug_irq, cpu).irq >= 0) { | |
9547689f | 130 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL); |
ee336e10 | 131 | per_cpu(xen_debug_irq, cpu).irq = -1; |
b85fffec KRW |
132 | kfree(per_cpu(xen_debug_irq, cpu).name); |
133 | per_cpu(xen_debug_irq, cpu).name = NULL; | |
ee336e10 KRW |
134 | } |
135 | if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) { | |
9547689f | 136 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq, |
53b94fdc | 137 | NULL); |
ee336e10 | 138 | per_cpu(xen_callfuncsingle_irq, cpu).irq = -1; |
b85fffec KRW |
139 | kfree(per_cpu(xen_callfuncsingle_irq, cpu).name); |
140 | per_cpu(xen_callfuncsingle_irq, cpu).name = NULL; | |
ee336e10 | 141 | } |
53b94fdc KRW |
142 | if (xen_hvm_domain()) |
143 | return; | |
144 | ||
ee336e10 | 145 | if (per_cpu(xen_irq_work, cpu).irq >= 0) { |
9547689f | 146 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL); |
ee336e10 | 147 | per_cpu(xen_irq_work, cpu).irq = -1; |
b85fffec KRW |
148 | kfree(per_cpu(xen_irq_work, cpu).name); |
149 | per_cpu(xen_irq_work, cpu).name = NULL; | |
ee336e10 | 150 | } |
53b94fdc | 151 | }; |
f87e4cac JF |
152 | static int xen_smp_intr_init(unsigned int cpu) |
153 | { | |
154 | int rc; | |
b85fffec | 155 | char *resched_name, *callfunc_name, *debug_name; |
f87e4cac JF |
156 | |
157 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
158 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
159 | cpu, | |
160 | xen_reschedule_interrupt, | |
9d71cee6 | 161 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
162 | resched_name, |
163 | NULL); | |
164 | if (rc < 0) | |
165 | goto fail; | |
9547689f | 166 | per_cpu(xen_resched_irq, cpu).irq = rc; |
b85fffec | 167 | per_cpu(xen_resched_irq, cpu).name = resched_name; |
f87e4cac JF |
168 | |
169 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
170 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
171 | cpu, | |
172 | xen_call_function_interrupt, | |
9d71cee6 | 173 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
174 | callfunc_name, |
175 | NULL); | |
176 | if (rc < 0) | |
177 | goto fail; | |
9547689f | 178 | per_cpu(xen_callfunc_irq, cpu).irq = rc; |
b85fffec | 179 | per_cpu(xen_callfunc_irq, cpu).name = callfunc_name; |
f87e4cac | 180 | |
ee523ca1 JF |
181 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
182 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
9d71cee6 | 183 | IRQF_PERCPU | IRQF_NOBALANCING, |
ee523ca1 JF |
184 | debug_name, NULL); |
185 | if (rc < 0) | |
186 | goto fail; | |
9547689f | 187 | per_cpu(xen_debug_irq, cpu).irq = rc; |
b85fffec | 188 | per_cpu(xen_debug_irq, cpu).name = debug_name; |
ee523ca1 | 189 | |
3b16cf87 JA |
190 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
191 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
192 | cpu, | |
193 | xen_call_function_single_interrupt, | |
9d71cee6 | 194 | IRQF_PERCPU|IRQF_NOBALANCING, |
3b16cf87 JA |
195 | callfunc_name, |
196 | NULL); | |
197 | if (rc < 0) | |
198 | goto fail; | |
9547689f | 199 | per_cpu(xen_callfuncsingle_irq, cpu).irq = rc; |
b85fffec | 200 | per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name; |
3b16cf87 | 201 | |
27d8b207 KRW |
202 | /* |
203 | * The IRQ worker on PVHVM goes through the native path and uses the | |
204 | * IPI mechanism. | |
205 | */ | |
206 | if (xen_hvm_domain()) | |
207 | return 0; | |
208 | ||
1ff2b0c3 LM |
209 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
210 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
211 | cpu, | |
212 | xen_irq_work_interrupt, | |
9d71cee6 | 213 | IRQF_PERCPU|IRQF_NOBALANCING, |
1ff2b0c3 LM |
214 | callfunc_name, |
215 | NULL); | |
216 | if (rc < 0) | |
217 | goto fail; | |
9547689f | 218 | per_cpu(xen_irq_work, cpu).irq = rc; |
b85fffec | 219 | per_cpu(xen_irq_work, cpu).name = callfunc_name; |
1ff2b0c3 | 220 | |
f87e4cac JF |
221 | return 0; |
222 | ||
223 | fail: | |
53b94fdc | 224 | xen_smp_intr_free(cpu); |
f87e4cac JF |
225 | return rc; |
226 | } | |
227 | ||
c7b75947 | 228 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
229 | { |
230 | int i, rc; | |
231 | ||
ea5b8f73 SS |
232 | if (xen_initial_domain()) |
233 | return; | |
234 | ||
235 | for (i = 0; i < nr_cpu_ids; i++) { | |
236 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
237 | if (rc >= 0) { | |
238 | num_processors++; | |
239 | set_cpu_possible(i, true); | |
240 | } | |
241 | } | |
242 | } | |
243 | ||
244 | static void __init xen_filter_cpu_maps(void) | |
245 | { | |
246 | int i, rc; | |
cf405ae6 | 247 | unsigned int subtract = 0; |
ea5b8f73 SS |
248 | |
249 | if (!xen_initial_domain()) | |
250 | return; | |
251 | ||
801fd14a SS |
252 | num_processors = 0; |
253 | disabled_cpus = 0; | |
e7986739 | 254 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 255 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
256 | if (rc >= 0) { |
257 | num_processors++; | |
4f062896 | 258 | set_cpu_possible(i, true); |
801fd14a SS |
259 | } else { |
260 | set_cpu_possible(i, false); | |
261 | set_cpu_present(i, false); | |
cf405ae6 | 262 | subtract++; |
4560a294 | 263 | } |
f87e4cac | 264 | } |
cf405ae6 KRW |
265 | #ifdef CONFIG_HOTPLUG_CPU |
266 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
267 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
268 | * have up to X, while nr_cpu_ids is greater than X. This | |
269 | * normally is not a problem, except when CPU hotplugging | |
270 | * is involved and then there might be more than X CPUs | |
271 | * in the guest - which will not work as there is no | |
272 | * hypercall to expand the max number of VCPUs an already | |
273 | * running guest has. So cap it up to X. */ | |
274 | if (subtract) | |
275 | nr_cpu_ids = nr_cpu_ids - subtract; | |
276 | #endif | |
277 | ||
f87e4cac JF |
278 | } |
279 | ||
a9e7062d | 280 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 281 | { |
f87e4cac JF |
282 | BUG_ON(smp_processor_id() != 0); |
283 | native_smp_prepare_boot_cpu(); | |
284 | ||
26a79995 | 285 | if (xen_pv_domain()) { |
5840c84b MR |
286 | if (!xen_feature(XENFEAT_writable_page_tables)) |
287 | /* We've switched to the "real" per-cpu gdt, so make | |
288 | * sure the old memory can be recycled. */ | |
289 | make_lowmem_page_readwrite(xen_initial_gdt); | |
60223a32 | 290 | |
7cde9b27 FZ |
291 | #ifdef CONFIG_X86_32 |
292 | /* | |
293 | * Xen starts us with XEN_FLAT_RING1_DS, but linux code | |
294 | * expects __USER_DS | |
295 | */ | |
296 | loadsegment(ds, __USER_DS); | |
297 | loadsegment(es, __USER_DS); | |
298 | #endif | |
299 | ||
26a79995 KRW |
300 | xen_filter_cpu_maps(); |
301 | xen_setup_vcpu_info_placement(); | |
302 | } | |
303 | /* | |
304 | * The alternative logic (which patches the unlock/lock) runs before | |
305 | * the smp bootup up code is activated. Hence we need to set this up | |
306 | * the core kernel is being patched. Otherwise we will have only | |
307 | * modules patched but not core code. | |
308 | */ | |
bf7aab3a | 309 | xen_init_spinlocks(); |
f87e4cac JF |
310 | } |
311 | ||
a9e7062d | 312 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
313 | { |
314 | unsigned cpu; | |
900cba88 | 315 | unsigned int i; |
f87e4cac | 316 | |
ed467e69 KRW |
317 | if (skip_ioapic_setup) { |
318 | char *m = (max_cpus == 0) ? | |
319 | "The nosmp parameter is incompatible with Xen; " \ | |
320 | "use Xen dom0_max_vcpus=1 parameter" : | |
321 | "The noapic parameter is incompatible with Xen"; | |
322 | ||
323 | xen_raw_printk(m); | |
324 | panic(m); | |
325 | } | |
2d9e1e2f JF |
326 | xen_init_lock_cpu(0); |
327 | ||
06d0b5d9 | 328 | smp_store_boot_cpu_info(); |
c7b75947 | 329 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
330 | |
331 | for_each_possible_cpu(i) { | |
332 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
333 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
334 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
335 | } | |
f87e4cac JF |
336 | set_cpu_sibling_map(0); |
337 | ||
338 | if (xen_smp_intr_init(0)) | |
339 | BUG(); | |
340 | ||
b78936e1 MT |
341 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
342 | panic("could not allocate xen_cpu_initialized_map\n"); | |
343 | ||
344 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
345 | |
346 | /* Restrict the possible_map according to max_cpus. */ | |
347 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 348 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 349 | continue; |
4f062896 | 350 | set_cpu_possible(cpu, false); |
f87e4cac JF |
351 | } |
352 | ||
7eb43a6d | 353 | for_each_possible_cpu(cpu) |
4f062896 | 354 | set_cpu_present(cpu, true); |
f87e4cac JF |
355 | } |
356 | ||
148f9bb8 | 357 | static int |
f87e4cac JF |
358 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
359 | { | |
360 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 361 | struct desc_struct *gdt; |
9976b39b | 362 | unsigned long gdt_mfn; |
f87e4cac | 363 | |
ce4b1b16 IM |
364 | /* used to tell cpu_init() that it can proceed with initialization */ |
365 | cpumask_set_cpu(cpu, cpu_callout_mask); | |
b78936e1 | 366 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
367 | return 0; |
368 | ||
369 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
370 | if (ctxt == NULL) | |
371 | return -ENOMEM; | |
372 | ||
c7b75947 JF |
373 | gdt = get_cpu_gdt_table(cpu); |
374 | ||
c7b75947 | 375 | #ifdef CONFIG_X86_32 |
5840c84b | 376 | /* Note: PVH is not yet supported on x86_32. */ |
c7b75947 | 377 | ctxt->user_regs.fs = __KERNEL_PERCPU; |
577eebea | 378 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
c7b75947 | 379 | #endif |
f87e4cac JF |
380 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); |
381 | ||
5840c84b | 382 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
a2ef5dc2 | 383 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
5840c84b | 384 | ctxt->flags = VGCF_IN_KERNEL; |
dacd45f4 KRW |
385 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ |
386 | ctxt->user_regs.ds = __USER_DS; | |
387 | ctxt->user_regs.es = __USER_DS; | |
5840c84b | 388 | ctxt->user_regs.ss = __KERNEL_DS; |
f87e4cac | 389 | |
dacd45f4 | 390 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 391 | |
dacd45f4 | 392 | ctxt->ldt_ents = 0; |
9976b39b | 393 | |
dacd45f4 | 394 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 395 | |
dacd45f4 KRW |
396 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
397 | make_lowmem_page_readonly(gdt); | |
398 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 399 | |
dacd45f4 KRW |
400 | ctxt->gdt_frames[0] = gdt_mfn; |
401 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 402 | |
dacd45f4 KRW |
403 | ctxt->kernel_ss = __KERNEL_DS; |
404 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 405 | |
c7b75947 | 406 | #ifdef CONFIG_X86_32 |
dacd45f4 KRW |
407 | ctxt->event_callback_cs = __KERNEL_CS; |
408 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
5840c84b MR |
409 | #else |
410 | ctxt->gs_base_kernel = per_cpu_offset(cpu); | |
c7b75947 | 411 | #endif |
dacd45f4 KRW |
412 | ctxt->event_callback_eip = |
413 | (unsigned long)xen_hypervisor_callback; | |
414 | ctxt->failsafe_callback_eip = | |
415 | (unsigned long)xen_failsafe_callback; | |
5840c84b MR |
416 | ctxt->user_regs.cs = __KERNEL_CS; |
417 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
dacd45f4 | 418 | } |
a2ef5dc2 MR |
419 | #ifdef CONFIG_XEN_PVH |
420 | else { | |
421 | /* | |
422 | * The vcpu comes on kernel page tables which have the NX pte | |
423 | * bit set. This means before DS/SS is touched, NX in | |
424 | * EFER must be set. Hence the following assembly glue code. | |
5840c84b | 425 | */ |
a2ef5dc2 | 426 | ctxt->user_regs.eip = (unsigned long)xen_pvh_early_cpu_init; |
5840c84b | 427 | ctxt->user_regs.rdi = cpu; |
a2ef5dc2 MR |
428 | ctxt->user_regs.rsi = true; /* entry == true */ |
429 | } | |
5840c84b | 430 | #endif |
dacd45f4 | 431 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); |
f87e4cac | 432 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir)); |
f87e4cac JF |
433 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) |
434 | BUG(); | |
435 | ||
436 | kfree(ctxt); | |
437 | return 0; | |
438 | } | |
439 | ||
148f9bb8 | 440 | static int xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 441 | { |
f87e4cac JF |
442 | int rc; |
443 | ||
3f85483b | 444 | common_cpu_up(cpu, idle); |
4461bbc0 | 445 | |
02889672 | 446 | xen_setup_runstate_info(cpu); |
f87e4cac | 447 | xen_setup_timer(cpu); |
2d9e1e2f | 448 | xen_init_lock_cpu(cpu); |
f87e4cac | 449 | |
2a442c9c PM |
450 | /* |
451 | * PV VCPUs are always successfully taken down (see 'while' loop | |
452 | * in xen_cpu_die()), so -EBUSY is an error. | |
453 | */ | |
454 | rc = cpu_check_up_prepare(cpu); | |
455 | if (rc) | |
456 | return rc; | |
c7b75947 | 457 | |
f87e4cac JF |
458 | /* make sure interrupts start blocked */ |
459 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
460 | ||
461 | rc = cpu_initialize_context(cpu, idle); | |
462 | if (rc) | |
463 | return rc; | |
464 | ||
f87e4cac JF |
465 | rc = xen_smp_intr_init(cpu); |
466 | if (rc) | |
467 | return rc; | |
468 | ||
f87e4cac JF |
469 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL); |
470 | BUG_ON(rc); | |
471 | ||
2a442c9c | 472 | while (cpu_report_state(cpu) != CPU_ONLINE) |
1207cf8e | 473 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 | 474 | |
f87e4cac JF |
475 | return 0; |
476 | } | |
477 | ||
a9e7062d | 478 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
479 | { |
480 | } | |
481 | ||
2737146b | 482 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 483 | static int xen_cpu_disable(void) |
d68d82af AN |
484 | { |
485 | unsigned int cpu = smp_processor_id(); | |
486 | if (cpu == 0) | |
487 | return -EBUSY; | |
488 | ||
489 | cpu_disable_common(); | |
490 | ||
491 | load_cr3(swapper_pg_dir); | |
492 | return 0; | |
493 | } | |
494 | ||
26fd1051 | 495 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 496 | { |
b12abaa1 | 497 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) { |
57b6b99b | 498 | __set_current_state(TASK_UNINTERRUPTIBLE); |
d68d82af AN |
499 | schedule_timeout(HZ/10); |
500 | } | |
54279552 | 501 | |
2a442c9c PM |
502 | if (common_cpu_die(cpu) == 0) { |
503 | xen_smp_intr_free(cpu); | |
504 | xen_uninit_lock_cpu(cpu); | |
505 | xen_teardown_timer(cpu); | |
506 | } | |
d68d82af AN |
507 | } |
508 | ||
148f9bb8 | 509 | static void xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
510 | { |
511 | play_dead_common(); | |
512 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
513 | cpu_bringup(); | |
466318a8 KRW |
514 | /* |
515 | * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down) | |
516 | * clears certain data that the cpu_idle loop (which called us | |
517 | * and that we return from) expects. The only way to get that | |
518 | * data back is to call: | |
519 | */ | |
520 | tick_nohz_idle_enter(); | |
d68d82af AN |
521 | } |
522 | ||
2737146b | 523 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 524 | static int xen_cpu_disable(void) |
2737146b AN |
525 | { |
526 | return -ENOSYS; | |
527 | } | |
528 | ||
26fd1051 | 529 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
530 | { |
531 | BUG(); | |
532 | } | |
533 | ||
26fd1051 | 534 | static void xen_play_dead(void) |
2737146b AN |
535 | { |
536 | BUG(); | |
537 | } | |
538 | ||
539 | #endif | |
f87e4cac JF |
540 | static void stop_self(void *v) |
541 | { | |
542 | int cpu = smp_processor_id(); | |
543 | ||
544 | /* make sure we're not pinning something down */ | |
545 | load_cr3(swapper_pg_dir); | |
546 | /* should set up a minimal gdt */ | |
547 | ||
086748e5 IC |
548 | set_cpu_online(cpu, false); |
549 | ||
f87e4cac JF |
550 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL); |
551 | BUG(); | |
552 | } | |
553 | ||
76fac077 | 554 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 555 | { |
76fac077 | 556 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
557 | } |
558 | ||
a9e7062d | 559 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
560 | { |
561 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
562 | } | |
563 | ||
f447d56d BG |
564 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
565 | int vector) | |
f87e4cac JF |
566 | { |
567 | unsigned cpu; | |
568 | ||
bcda016e | 569 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
570 | xen_send_IPI_one(cpu, vector); |
571 | } | |
572 | ||
bcda016e | 573 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
574 | { |
575 | int cpu; | |
576 | ||
f447d56d | 577 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
578 | |
579 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 580 | for_each_cpu(cpu, mask) { |
3b16cf87 | 581 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 582 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
583 | break; |
584 | } | |
585 | } | |
586 | } | |
587 | ||
a9e7062d | 588 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 589 | { |
f447d56d | 590 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 591 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
592 | } |
593 | ||
f447d56d BG |
594 | static inline int xen_map_vector(int vector) |
595 | { | |
596 | int xen_vector; | |
597 | ||
598 | switch (vector) { | |
599 | case RESCHEDULE_VECTOR: | |
600 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
601 | break; | |
602 | case CALL_FUNCTION_VECTOR: | |
603 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
604 | break; | |
605 | case CALL_FUNCTION_SINGLE_VECTOR: | |
606 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
607 | break; | |
1ff2b0c3 LM |
608 | case IRQ_WORK_VECTOR: |
609 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
610 | break; | |
6efa20e4 KRW |
611 | #ifdef CONFIG_X86_64 |
612 | case NMI_VECTOR: | |
613 | case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */ | |
614 | xen_vector = XEN_NMI_VECTOR; | |
615 | break; | |
616 | #endif | |
f447d56d BG |
617 | default: |
618 | xen_vector = -1; | |
619 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
620 | vector); | |
621 | } | |
622 | ||
623 | return xen_vector; | |
624 | } | |
625 | ||
626 | void xen_send_IPI_mask(const struct cpumask *mask, | |
627 | int vector) | |
628 | { | |
629 | int xen_vector = xen_map_vector(vector); | |
630 | ||
631 | if (xen_vector >= 0) | |
632 | __xen_send_IPI_mask(mask, xen_vector); | |
633 | } | |
634 | ||
635 | void xen_send_IPI_all(int vector) | |
636 | { | |
637 | int xen_vector = xen_map_vector(vector); | |
638 | ||
639 | if (xen_vector >= 0) | |
640 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
641 | } | |
642 | ||
643 | void xen_send_IPI_self(int vector) | |
644 | { | |
645 | int xen_vector = xen_map_vector(vector); | |
646 | ||
647 | if (xen_vector >= 0) | |
648 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
649 | } | |
650 | ||
651 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
652 | int vector) | |
653 | { | |
654 | unsigned cpu; | |
655 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 656 | int xen_vector = xen_map_vector(vector); |
f447d56d | 657 | |
1db01b49 | 658 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
659 | return; |
660 | ||
661 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
662 | if (this_cpu == cpu) | |
663 | continue; | |
664 | ||
1db01b49 | 665 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
666 | } |
667 | } | |
668 | ||
669 | void xen_send_IPI_allbutself(int vector) | |
670 | { | |
1db01b49 | 671 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
672 | } |
673 | ||
f87e4cac JF |
674 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
675 | { | |
f87e4cac | 676 | irq_enter(); |
3b16cf87 | 677 | generic_smp_call_function_interrupt(); |
1b437c8c | 678 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
679 | irq_exit(); |
680 | ||
f87e4cac JF |
681 | return IRQ_HANDLED; |
682 | } | |
683 | ||
3b16cf87 | 684 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 685 | { |
3b16cf87 JA |
686 | irq_enter(); |
687 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 688 | inc_irq_stat(irq_call_count); |
3b16cf87 | 689 | irq_exit(); |
f87e4cac | 690 | |
3b16cf87 | 691 | return IRQ_HANDLED; |
f87e4cac | 692 | } |
a9e7062d | 693 | |
1ff2b0c3 LM |
694 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
695 | { | |
696 | irq_enter(); | |
697 | irq_work_run(); | |
698 | inc_irq_stat(apic_irq_work_irqs); | |
699 | irq_exit(); | |
700 | ||
701 | return IRQ_HANDLED; | |
702 | } | |
703 | ||
b53cedeb | 704 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
705 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
706 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
707 | .smp_cpus_done = xen_smp_cpus_done, |
708 | ||
d68d82af AN |
709 | .cpu_up = xen_cpu_up, |
710 | .cpu_die = xen_cpu_die, | |
711 | .cpu_disable = xen_cpu_disable, | |
712 | .play_dead = xen_play_dead, | |
713 | ||
76fac077 | 714 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
715 | .smp_send_reschedule = xen_smp_send_reschedule, |
716 | ||
717 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
718 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
719 | }; | |
720 | ||
721 | void __init xen_smp_init(void) | |
722 | { | |
723 | smp_ops = xen_smp_ops; | |
c7b75947 | 724 | xen_fill_possible_map(); |
a9e7062d | 725 | } |
99bbb3a8 SS |
726 | |
727 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
728 | { | |
729 | native_smp_prepare_cpus(max_cpus); | |
730 | WARN_ON(xen_smp_intr_init(0)); | |
731 | ||
99bbb3a8 | 732 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
733 | } |
734 | ||
148f9bb8 | 735 | static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle) |
99bbb3a8 SS |
736 | { |
737 | int rc; | |
2a442c9c PM |
738 | |
739 | /* | |
740 | * This can happen if CPU was offlined earlier and | |
741 | * offlining timed out in common_cpu_die(). | |
742 | */ | |
743 | if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) { | |
744 | xen_smp_intr_free(cpu); | |
745 | xen_uninit_lock_cpu(cpu); | |
746 | } | |
747 | ||
fc78d343 CA |
748 | /* |
749 | * xen_smp_intr_init() needs to run before native_cpu_up() | |
750 | * so that IPI vectors are set up on the booting CPU before | |
751 | * it is marked online in native_cpu_up(). | |
752 | */ | |
753 | rc = xen_smp_intr_init(cpu); | |
754 | WARN_ON(rc); | |
755 | if (!rc) | |
756 | rc = native_cpu_up(cpu, tidle); | |
1fb3a8b2 KRW |
757 | |
758 | /* | |
759 | * We must initialize the slowpath CPU kicker _after_ the native | |
760 | * path has executed. If we initialized it before none of the | |
761 | * unlocker IPI kicks would reach the booting CPU as the booting | |
762 | * CPU had not set itself 'online' in cpu_online_mask. That mask | |
763 | * is checked when IPIs are sent (on HVM at least). | |
764 | */ | |
765 | xen_init_lock_cpu(cpu); | |
99bbb3a8 SS |
766 | return rc; |
767 | } | |
768 | ||
99bbb3a8 SS |
769 | void __init xen_hvm_smp_init(void) |
770 | { | |
3c05c4be SS |
771 | if (!xen_have_vector_callback) |
772 | return; | |
99bbb3a8 SS |
773 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; |
774 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
775 | smp_ops.cpu_up = xen_hvm_cpu_up; | |
2a442c9c | 776 | smp_ops.cpu_die = xen_cpu_die; |
99bbb3a8 SS |
777 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; |
778 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
26a79995 | 779 | smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu; |
99bbb3a8 | 780 | } |